High speed integrated circuits formed on a crystalline semiconductor wafer have ultra shallow semiconductor junctions formed by ion implanting dopant impurities into source and drain regions. The implanted dopant impurities are activated by a high temperature anneal step which causes a large proportion of the implanted atoms to become substitutional in the crystalline semiconductor lattice. Such a post-ion implantation anneal step are done by a rapid thermal process (RTP)-employing powerful lamps that heat the entire wafer volume to a very high temperature for a short time (e.g., a rate-of-rise of about 100-200 degrees C. per second and an intial rate-of-fall of 50-100 degrees C. per second). The heating duration must be short to avoid degrading the implanted junction definition by thermally induced diffusion of the dopant impurities from their implanted locations in the semiconductor wafer. This RTP approach is a great improvement over the older post-ion implant anneal technique of heating the wafer in a furnace for a long period of time. RTP using lamps is effective because the time response of the heat source (the lamp filament) is short in contrast to the furnace annealing step in which the heater response time is very slow. The high temperature, short duration heating of the RTP method favors the activation of implanted impurities while minimizing thermally induced diffusion.
An improved anneal is done by a flash lamp anneal process employing powerful flash lamps that heat the surface (only) of the entire wafer to a very high temperature for a very short time (e.g., a few milliseconds). The heating duration must be short to avoid degrading the implanted junction definition by thermally induced diffusion of the dopant impurities from their implanted locations in the semiconductor wafer. This flash approach is an improvement over the RTP approach, because the bulk of the wafer acts as a heat sink and permits rapid cooling of the hot wafer surface. High speed anneal using flashlamps is more effective because the heating is confined to the surface of the wafer, in contrast to the RTP annealing step in which the entire volume of the wafer is heated to approximately the same anneal temperature. The short duration at high temperature of the flash method minimizes thermally induced diffusion. However, it is difficult to achieve thermal uniformity over the entire wafer. Greater thermal non-uniformity within wafer creates significant amount of mechanical stress, resulting in wafer breakage and limits the highest operating temperature to approximately 1150° C. for anneal using flash lamps. The surface temperature during flashlamp annealing is determined by the intensity and pulse duration of flashlamps, which are difficult to control in a repeatable manner from one wafer to the next
One problem with RTP is that as device size decreases to 65 nanometers (nm) and below, the minimal thermal diffusion caused by RTP or flash heating becomes significant relative to the device size, despite the short duration of the RTP or flash heating. Another problem is that the degree of activation of the implanted dopant impurities is limited by the maximum temperature of the RTP or flash process. Heating the entire wafer volume in the RTP process above the maximum temperature (e.g., 1100 degrees C.) can create mechanical stresses in the wafer that cause lattice defects and wafer breakage in extreme cases. Limiting the wafer temperature to a maximum level (e.g., 1100 degrees C.) prevents such breakage, but unfortunately limits the proportion of implanted (dopant) atoms that are activated (i.e., that become substitutional in the semiconductor crystalline lattice). Limiting the dopant activation limits sheet conductivity and limits device speed. This problem becomes more significant as device size is reduced below 65 nm (e.g., down to 45 nm).
In order to raise the level of dopant activation beyond that achieved by RTP or flash annealing, laser annealing has been introduced as a replacement for RTP. One type of laser that has been used is a CO2 laser having an emission wavelength of 10.6 microns. This laser produces a narrow cylindrical beam, which must be raster-scanned across the entire wafer surface. In order to decrease the surface reflectivity at 10.6 microns, the beam is held at an acute angle relative to the wafer surface. Since the CO2 laser wavelength corresponds to a photon energy less than the bandgap of silicon, the silicon must be pre-heated to populate the conduction band with free carriers in order to facilitate the absorption of 10.6 micron photons through free carrier absorption. A fundamental problem is that the absorption at 10.6 microns is pattern-dependent because it is affected by the dopant impurities (which among other factors, determines the local free carrier concentration), so that the wafer surface is not heated uniformly. Also, conductive or metallic features on the wafer are highly reflective at the 10.6 micron laser wavelength, so that this process may not be useful in the presence of conductive thin film features.
The post-implant anneal step has been performed with short wavelength pulsed lasers (the short wavelength corresponding to a photon energy greater than the bandgap of silicon). While the surface heating is extremely rapid and shallow, such pulsed lasers bring the semiconductor crystal to its melting point, and therefore the heating must be restricted to an extremely shallow depth, which reduces the usefulness of this approach. Typically, the depth of the heated region does not extend below the depth of the ultra-shallow junctions (about 200 Angstroms).
The foregoing problems have been overcome by employing an array of diode lasers whose multiple parallel beams are focused along a narrow line (e.g., about 300 microns wide) having a length on the order of the wafer diameter or radius. The diode lasers have a wavelength of about 810 nm. This wavelength corresponds to a photon energy in excess of the bandgap energy of the semiconductor crystal (silicon), so that the laser energy excites electron transitions between the valence and conduction bands, which subsequently release the absorbed energy to the lattice and raises the lattice temperature. The narrow laser beam line is scanned transversely across the entire wafer surface (e.g., at a rate of about 300 mm/sec), so that each point on the wafer surface is exposed for a very short time (e.g., about 1 millisec). This type of annealing is disclosed in U.S. Patent Publication No. US 2003/0196996A1 (Oct. 23, 2003) by Dean C. Jennings et al. The wafer is scanned much more quickly by the wide thin beam line than by the pencil-like beam of a single laser spot, so that productivity is much greater, approaching that of RTP. But, unlike RTP, only a small portion of the wafer surface is heated, so that the stress is relieved in the remaining (bulk) portion of the wafer, allowing the peak temperature to be increased above the maximum RTP temperature (e.g., to about 1250-1300 degrees C.). The entire wafer volume may also be preheated during the laser scanning anneal in order to improve the annealing characteristics. The maximum preheated temperature is dictated by the technology nodes, process requirements, compatibility with semiconductor materials, etc. As a result, dopant activation is much higher, so that sheet resistivity is lower and device speed is higher. Each region of the wafer surface reaches a temperature range of about 1250-1300 degrees C. for about 50 microsec. The depth of this region is about 10-20 microns. This extends well-below the ultra-shallow semiconductor junction depth of about 200 Angstroms.
The wafer surface must be heated above a minimum temperature (e.g., 1250 degrees C.) in order to achieve the desired degree of activation of the implanted (dopant) atoms. The elevated temperature is also required to anneal other lattice damage and defects caused by any preceeding implant or thermal steps, in order to improve the electrical characteristics of the junctions such as their electrical conductivity and leakage. The wafer surface must be kept below a maximum temperature (e.g., 1350 degrees C.) in order to avoid the melting temperature of the semiconductor crystal (e.g., crystalline or polycrystalline silicon). In order to uniformly heat the entire wafer surface within this desired temperature range, the optical absorption of the wafer surface must be uniform across the wafer, and the surface temperature in the illuminated portion of the wafer surface must be accurately monitored while the laser beam line is scanned across the wafer (to enable precise temperature control). This is accomplished by measuring the emission of light by the heated portion of the wafer surface (usually of a wavelength different from that of the laser light source), and the measurement must be uniformly accurate. As employed in this specification, the term “optical” is meant to refer to any wavelength of a light or electromagnetic radiation emitted from a light source (such as a laser) that is infrared or visible or ultraviolet or emitted from the heated wafer surface.
The problem is that the underlying thin film structures formed on the wafer surface present different optical absorption characteristics and different optical emissivities in different locations on the wafer surface. This makes it difficult if not impossible to attain uniform anneal temperatures across the wafer surface and uniformly accurate temperature measurements across the wafer surface. This problem can be solved by depositing a uniform optical absorption layer over the entire wafer surface that uniformly absorbs the laser radiation and then conducts the heat to the underlying semiconductor wafer. Such a film must withstand the stress of heating during the laser anneal step without damage or separation, and must be selectively removable after the laser anneal step with respect to underlayers and must not contaminate or damage the underlying semiconductor wafer or thin film features. Further, the absorber film must attain excellent step coverage (high degree of conformality) over the underlying thin film features. One advantage of such a film is that lateral heat conduction in the film can mask non-uniformities in the light beam. This approach has been attempted but has been plagued by problems. One type of absorber layer consists of alternating metal and dielectric layers that form an anti-reflective coating. The different layers in this type of absorber material tend to fuse together under the intense heat of the laser beam, and become difficult to remove following the laser anneal step or contaminate underlying layers with metal.
A better approach used in the present invention is to employ an absorber layer that can be deposited by plasma enhanced chemical vapor deposition (PECVD). As disclosed in U.S. patent application Ser. No. 10/679,189, filed Oct. 3, 2003 by Luc Van Autryve et al. entitled “Absorber Layer for DSA Processing” and assigned to the present assignee, the PECVD-deposited absorber layer may be amorphous carbon. One advantage of amorphous carbon is that it is readily and selectively (with respect to underlayers of other materials) removed by oxidation in a plasma process or a downstream oxidation process employing radicals, at a wafer temperature less than 400 C. Another advantage is that carbon is generally compatible with semiconductor plasma processes and therefore does not involve contamination, so long as excessive implantation does not occur. One problem is that the deposited layer is vulnerable to cracking or peeling under the high temperatures of the laser anneal step, unless the layer is deposited at a very high temperature (e.g., 550 degrees C.). (The tendency or resistance to such cracking, peeling or separation of the deposited layer from the underlying layer in response to high temperature or high temperature gradients is generally referred to in this specification as the thermal or thermal-mechanical properties of the deposited layer.) Also the thermal budget (time and temperature) associated with this PECVD deposition process caused dopants to form clusters which are difficult to dissolve with the subsequent laser anneal step, particularly for feature sizes below 65 nm (such as feature sizes of about 45 nm). Attempting to solve this problem by reducing the wafer temperature (e.g., to 400 degrees C.) during PECVD deposition of the absorber layer material creates two problems. First, the thermal properties of the deposited layer are such that it will fail (by cracking, peeling or separation from the wafer) during the laser annealing step. Secondly, the deposited layer that is produced is transparent or has insufficient optical absorption. Another problem encountered with this absorber layer is that it has poor step coverage. We have observed that the PECVD 550 degree absorber layer can have very large voids in the vicinity of pronounced steps in the underlying layer or thin film structures sizes below 65 nm.
We feel that failure of the absorber layer (e.g., by peeling or cracking) arises from a lack of high quality chemical bonds (between the underlying layer and the deposited material) capable of withstanding the stress of being rapidly heated to 1300 degrees C. during the laser anneal step. We feel that, in order to improve the thermal properties of the deposited layer, achieving such high quality bonds at low wafer temperature requires high ion energies during the PECVD process. Such high ion energies are not readily attainable in conventional PECVD reactors. We feel that poor step coverage by the absorber layer or amorphous carbon layer is the result of the inability of a conventional PECVD or HDPCVD reactor to provide an intermediate range of ionization (ion-to-radical ratio) with an adequate level of energetic ion bombardment. These inadequacies arise, in part, because such conventional PECVD and HDPCVD reactors cannot operate within a wide intermediate range of source power coupling (to generate plasma electrons), chamber pressure and wafer voltage. Indeed, the different types of conventional PECVD and HDPCVD reactors tend to operate at either very high or very low ranges of source power coupling (to generate plasma electrons), chamber pressure and wafer voltage. Conventional PECVD reactors employ capacitively-coupled RF source power at relatively high-pressure, resulting in a very low range of ionization (ion-to-radical ratio) with an inadequate level of energetic ion bombardment (and no separate control of voltage or energy). This is due to the inefficient source power coupling (to generate plasma electrons) and the damping of ion energies by collisions with neutrals at high pressure. Even if separate RF biasing of the wafer is added, the damping of ion energies by collisions with neutrals at high pressure limits the voltage and energy range to a low range. Conversely, conventional HDPCVD reactors typically employ inductively-coupled RF source power at very low pressure. This type of plasma source typically initiates the plasma capacitively, and then has a high power threshold to transition to inductively coupled power mode. Once the power coupled is above this threshold and the source is operating in an inductive mode, the source power coupling is highly efficient and the minimum possible plasma density and range of ionization (ion-to-radical ratio) is very high. The separate RF wafer bias is coupled to the relatively dense plasma, which presents a very low electrical impedance load. The resultant RF bias power required to produce energetic ion bombardment is very high (>>10 kW for >2kV). High energies are not generally attainable due to practical RF delivery system limitations (RF generators, matching networks, and feed structures). Most of the bias power (e.g., ˜80%) is dissipated as heat on the wafer. It is very difficult to remove the heat at low pressure at an adequate rate to maintain low wafer temperature (<400 deg. C. or lower). Finally, both capacitively-coupled PECVD and inductively-coupled HDPCVD reactors may have power coupling drift (with on-time) issues when used with carbon chemistry when depositing absorbing or semiconducting films (on RF windows or insulators). The need (fulfilled by the toroidal plasma CVD reactor and process described in detail below) is for a reactor capable of providing ionization ratios in a wide intermediate range together with an adequate level of energetic ion bombardment in all cases, through an ability to operate in a wide intermediate range of source power coupling and level, wafer voltage and chamber pressure. The toroidal plasma CVD reactor does not exhibit power coupling drift when used with carbon chemistry when depositing absorbing or semiconducting films. This is because the toroidal plasma CVD reactor is already conducting (metal), having only very thin, isolated DC breaks, which do not accumulate much deposition and are easily in-situ plasma cleaned.
One type of conventional PECVD reactor is a capacitively coupled plasma reactor having a pair of closely-spaced parallel plate electrodes across which RF plasma source power is applied. Such a capacitively coupled reactor typically is operated at high chamber pressure (2-10 Torr). High pressure and close-spacing (relative to electrode radius) are employed to maximize deposition rate on the wafer, and to minimize deposition outside the process region. The plasma source power couples to both electrons in the bulk plasma and to ions in the plasma sheaths. The voltage across the electrodes is typically relatively low (less than 1 KVpp at source power of several kW for 300 mm wafer) and the plasma sheath is very collisional, so that the ion energy is typically low. This type of reactor produces a very low ion-to-neutral population ratio and ion-to-radical ratio, so that the ion flux is low, which probably increases the ion energy level or wafer temperature required to obtain the requisite high quality bonds between the deposited and underlying materials. However, because of the low inter-electrode voltage and the high loss of ion energy in the collisional sheath, it is very difficult to generate the ion energy distribution required for high quality bonds.
Another type of conventional PECVD reactor is an inductively coupled high plasma density CVD (HPDCVD) reactor in which RF source power is applied to an inductive antenna. The reactor must be operated at a low chamber pressure (e.g., 5-10 milliTorr) and high plasma source power level, because of the high minimum induced electric field required to maintain the inductively coupled plasma mode, which in turn produces a high plasma density. The degree of ionization (ratio of ion-to-neutral density) produced in this reactor is confined to a range of very high values (four or five orders of magnitude greater than that of the capacitive reactor discussed above), because a large amount of RF source power is required to sustain the inductively coupled mode and because the RF induced electric field couples directly to electrons in the bulk plasma. This contrasts with a capacitively coupled plasma in which the RF electric field less efficiently couples to electrons indirectly by displacement across the plasma sheath or through plasma sheath oscillations. As a result, plasma density and conductivity is very high, making it difficult to generate a high wafer voltage at practical bias power levels (since the wafer voltage is loaded down through the highly conductive plasma). As a result, high ion energies cannot be attained without applying excessive amounts of RF bias power to the wafer. This could overheat the wafer and perhaps destroy the ultra shallow junction definition in the underlying semiconductor crystal lattice (by thermal diffusion). Typically, for a 300 mm wafer, a wafer voltage of 1-2 kV peak-to-peak would require RF bias power of about 10 kWatts. Cooling the wafer to maintain ultra-shallow junction definition is difficult at high bias power, and even higher bias voltage (than 1-2 kV) and thus higher power is desired for best film properties. RF power delivery systems >10 kW are very expensive and have limited availability.
Another problem with the HDPCVD reactor is that a large non-conductive window must be provided in the chamber ceiling through which the plasma source power may be inductively coupled from the coil antenna. This prevents the use of a conductive showerhead directly overlying the wafer, which limits gas distribution uniformity at the wafer and RF bias ground reference uniformity over the wafer. Moreover, coupling of source power into the chamber may be effectively reduced or even blocked if the reactor is employed to deposit a non-insulating material on the wafer, since that same material will also accumulate on the dielectric window during processing, creating a conductive shield or semi-conductive attenuator to the RF power. The temperature of a non-conductive surface, such as the dielectric window of the HDPCVD reactor, cannot be effectively controlled, so that deposition during processing and post-process cleaning of the reactor interior is more difficult. A related problem in both types of reactors is that plasma source power seeks a ground return from any available conductive surface in the chamber, so that process control is hampered by electrical changes due to deposition of by-products on the chamber surfaces. With both dielectric and metallic materials constituting the chamber surfaces, removal of deposited plasma by-products after processing may be difficult or may involve undue wear of chamber parts. This may be circumvented by employing disposable shields or process kits to prevent deposition on chamber surfaces. However, such disposable shields cannot provide good RF ground reference nor be thermally controlled with any precision.
In summary, the conventional reactors are either confined to a narrow low chamber pressure window (in the case of the HDPCVD reactor) or a narrow high chamber pressure window (in the case of the capacitively coupled reactor). Neither chamber can achieve a high ion energy, either because the sheath is highly collisional (in the capacitively coupled reactor) or because the plasma is highly conductive (in the HDPCVD reactor). Also, they are confined to either a narrow high degree-of-ionization regime (the HDPCVD reactor) or a narrow low degree-of-ionization regime (the capacitively coupled reactor). Moreover, both types of reactors are susceptible to wide deviations in performance whenever they are used for deposition of non-insulating materials, since the accumulation of non-insulating materials across electrode boundaries in a capacitively coupled reactor or on the dielectric window of an inductively coupled reactor will distort or inhibit the coupling of RF source power into the chamber. What is needed is a deposition process carried out at a very low temperature (e.g., room temperature up to several hundred degrees C.) for forming an optical absorber layer having such high quality bonds with the underlying layers (including the semiconductor lattice) that it is impervious to mechanical failure or separation during the laser annealing step. The process should have a wide source power window, a wide degree-of-ionization window in an intermediate range, a wide wafer voltage (bias power) window with wide ion energy window, and a wide wafer temperature window.